|
Abstract:
|
Crime surveys are not designed to produce reliable estimates of crime in small geographic areas, and aggregates of police-recorded crimes are affected by measurement error driven by under-reporting. Thus, police forces and researchers ask for small area estimates. Small area estimation methods, borrowing strength from auxiliary data (e.g., Census), produce more efficient estimates than direct estimators based on sample information only. We produce estimates of collective efficacy and crime prevalence for London small areas using the Metropolitan Police Service Public Attitudes Survey 2015-17 data. Collective efficacy, which refers to how members of a community control individual behaviors, is based on a latent continuous variable constructed from a latent variable model. This is highly associated with crime. Crime prevalence refers to the proportion of citizens that have been victim of crime in the last year. The efficiency of small area estimates is improved by taking into account for correlations between these variables via multivariate models. Here we adopt a joint mixed-effect model for variables measured on different scales, which performs better than the univariate approach.
|
